THEORY OF MATROIDS AND APPLICATIONS
JORGE RAMÍREZ ALFONSÍN
Abstract. A matroid is a combinatorial structure that can be defined by keeping
the main ‘set properties’ of the linear dependency in vector spaces. Matroids satisfy several equivalent axioms and have a fundamental notion of duality giving the
right setting to study a number of classical problems. In this notes, we give a basic
introduction of the theory of matroids as well as some applications.
Contents
1. Introduction
2. Axioms
2.1. Independents
2.2. Circuits
2.3. Bases
2.4. Rank
2.5. Closure
2.6. Greedy Algorithm
3. Applications : combinatorial optimisation
3.1. Transversal matroid
3.2. Assigment Problem
4. Duality
4.1. Graphic matroids
4.2. Minors
5. Representable matroids
6. Regular matroids
7. Union of matroids
8. Tutte Polynomial
8.1. General properties
8.2. Chromatic Polynomial
8.3. Chromatic polynomial with defect
8.4. Flow Polynomial
8.5. Ehrhart polynomial
8.6. Acyclic and totally cyclic orientations
References
Partially supported by the ANR TEOMATRO grant ANR-10-BLAN 0207.
1
1
2
2
2
3
4
4
5
5
5
5
6
7
7
9
10
11
12
12
12
13
13
13
14
14
2
JORGE RAMÍREZ ALFONSÍN
1. Introduction
The word of matroid appears for the first time in the fundamental paper due to Whitney in 1935 [16]. In this paper, the structure of a matroid was introduced as the
set abstraction of the dependency relations between the column vectors of a matrix
(explaining the suffix ‘oid’ indicating that is a structure coming from a matrix). The
applications of matroids have their origins in the area of combinatorics (graphs, discrete optimisation, polytopes, etc.). For the last thirty years, many other applications
of matroid have appeared in many other differents areas as algebraic geometry, stratification of grassmanniennes, discrete geometry, etc. We recommend the reader the book
[9] for further details on matroid theory.
2. Axioms
Matroids can be characterized by various different equivalent axioms.
2.1. Independents. A matroid M is an ordered couple (E, I) where E is a finite set
(E = {1, . . . , n}) and I is a collection of subsets of E verifying the following
(I1) ∅ ∈ I,
(I2) If I1 ∈ I and I2 ⊂ I1 then I2 ∈ I,
(I3) (augmentation property) If I1 , I2 ∈ I and |I1 | < |I2 | then there exist e ∈ I2 \I1
such that I1 ∪ e ∈ I.
The members of I are called the independents of M . A subset of E that is not in I is
called dependent.
Example 2.1. Let {e1 , . . . , en } be the set of column vectors of a matrix with coefficients
in a field F. Let I be the collection of all the subsets of {i1 , . . . , im } ⊆ {1, . . . , n} = E
such that the set of columns {ei1 , . . . , eim } are linearly independente on F. Then, (E, I)
is a matroid.
A matroid that is isomorpic to a matroid obtained from a matrix over a field F is called
representable ou linear over F.
Example 2.2. Let A be the following matrix with coefficients in R
1 2 3 4 5
1 0 0 1 1
A=
0 1 0 0 1
We have that I(M ) = {∅, {1}, {2}, {3}, {4}, {1, 2}, {1, 3}, {2, 4}, {3, 4}, {2, 3}}.
2.2. Circuits. A subset X ⊆ E is said to be minimal dependent if any subset of X,
different from X, is independent. A minimal dependent subset of a matroid M is called
a circuit of M . We denote by C the set of circuits of a matroid. We observe that I
can be determined by C (the membres of I are all the subsets of E not containing a
member of C). A circuit of cardinality one is called loop.
We have that C is a set of circuits of a matroid on E if and only if C verifies the
following properties :
(C1) ∅ 6∈ C,
(C2) C1 , C2 ∈ C and C1 ⊆ C2 then C1 = C2 ,
(C3) (elimination property) If C1 , C2 ∈ C, C1 6= C2 and e ∈ C1 ∩ C2 then there exists
C3 ∈ C such that C3 ⊆ {C1 ∪ C2 }\{e}.
THEORY OF MATROIDS AND APPLICATIONS
3
Example 2.3. In the graph G = (V, E), we call cycle the set of edges in a cycle of G,
and simple cycle if such a set is minimal by inclusion. Here, we consider only simple
cycles. Let G = (V, E) be a graph and let C be the set of cycles of G. Then, C is the
set of circuits of a matroid on E, denoted by M (G). A matroid obtained on this way
is called graphic matroid.
e5
e4
e1
e3
e2
Figure 1. Graph with 3 verticess
Remark 2.4. There is not the notion of vertex in a matroid. The matroid associated
to a graph do not determine, in general, the graph. For instance, any connected graph
without cycle has the same matroid. The associated matroid determine the graph if
and only if the graph is 3-connected.
Example 2.5. Let M (G) be the graphic matroid obtained from graph G in Figure
1, and let A be the matrix over R given in Exemple 2.2. We can verify that M (G) is
isomorphic to M (A) (under the bijection ei → i).
Remark 2.6. It turns out that a graphic matroid is always linear. Let G = (V, E) be
a graph and {xi , i ∈ V } the canonique bases of a vector space over an arbitrary field.
We can verify that, as in Example 2.5, the graph G = (V, E) is associated to the same
matroid as the set of vectors xi − xj pour (i, j) ∈ E.
Example 2.7. Let M (G0 ) be the graphic matroid obtained from graph G0 in Figure
2. Let A0 be the matrix obtained by following the construction of Remark 2.6.
x1 x2 x3 x4 
1
1
0
0
0

0
1
0 
A =  −1

 0 −1 −1
1 
0
0
0 −1

We can check that M (G0 ) is isomorphic to M (A0 ) (under the bijection xi → i). Notice
that the cycle formed by the edges a = {1, 2}, b = {1, 3} et c = {2, 3} in the graph G0
correspond to the linear dependency x2 − x1 = x3 .
2.3. Bases. A base of a matroid is a independ set maximal by inclusion. All the
bases of a matroid have the same cardinality (the same number of elements). Indeed,
otherwise, we would have two bases B1 , B2 with |B1 | < |B2 | and so, by (I3) there
exist x ∈ B2 \B1 such that B1 ∪ x ∈ I which is a contradiction since B1 is a maximal
independent.
The collection B of bases verify the following conditions
(B1) B 6= ∅.
4
JORGE RAMÍREZ ALFONSÍN
1
a
b
c
2
d
3
4
Figure 2. Graph G0
(B2) (exchange property) B1 , B2 ∈ B and x ∈ B1 \B2 then there exist y ∈ B2 \B1
such that (B1 \x) ∪ y ∈ B.
If I is the collection of subsets contained in one of the members of B then (E, I) is a
matroid.
Remark 2.8. If G is a connected graph then the bases of M (G) correspond to the set
of all spanning trees of G.
2.4. Rank. Let M = (E, I) and X ⊆ E. The rank of X, denoted by rM (X), is the
cardinality of the largest independent contained in X, that is,
rM (X) = max{|Y | : Y ⊆ X, Y ∈ I}.
Equivalently, we define the set I\X = {I ⊆ X|I ∈ I}. Then, (X, I\X) is a matroid,
denoted by M |X and called restriction of M to X. The rank rM (X) of X is the
cardinality of a base in M |X .
It can be proved that r = rM is the rank function of a matroid (E, I) where
I = {I ⊆ E : r(I) = |I|},
if and only if r verify the following conditions
(R1) 0 ≤ r(X) ≤ |X|, for all X ⊆ E,
(R2) r(X) ≤ r(Y ), for all X ⊆ Y ,
(R3) (sub-modularity) r(X ∪ Y ) + r(X ∩ Y ) ≤ r(X) + r(Y ) for all X, Y ⊂ E.
2.5. Closure. The closure of a set X in M is defined by
cl(X) = {x ∈ E|r(X ∪ x) = r(X)}.
It can be proved that the function cl : P(E) → P(E) is the closure function of a
matroid if and only if cl verify the following conditions
(CL1) (extensivity) If X ⊆ E then X ⊆ cl(X).
(CL2) (increasing) If X ⊆ Y ⊆ E then cl(X) ⊆ cl(Y ).
(CL3) (indempotent) If X ⊆ E then cl(cl(X)) = cl(X).
(CL4) (exchange property) If X ⊆ E, x ∈ E and y ∈ cl(X ∪ x) − cl(X) then x ∈
cl(X ∪ y).
Let X ⊂ E, cl(X) is also called flat of X. X is said to be closed if X = cl(X). The set
E is a closed set of rank rM . The rank 0 closed sets are formed by the loops of M . A
closed set of rank 1 or point is the class of parallel elements. A matroid such that ∅ is
a closed set and all its points contain only one element, is called simple. A hyperplan
is a closed set of rank rM − 1.
THEORY OF MATROIDS AND APPLICATIONS
5
Example 2.9. Let M (G0 ) be a graphic matroid obtained from graph G0 in Figure 2.
It can be verified that : r({a, b, c}) = r({c, d}) = r({a, d}) = 2 and cl{a, b} = {a, b, c}.
2.6. Greedy Algorithm. Let I be a set of subsets of E verifying (I1)Pand (I2). Let
w : E → R (w(e) is said to be the weight of element e). Let w(X) = x∈X w(x) for
X ⊆ E, and w(∅) = 0.
An optimization problem consist to find a maximal set B of I with maximal weight.
Greedy algorithm for (I, w)
X0 = ∅
j=0
While there exist e ∈ E\Xj such that Xj ∪ {e} ∈ I do
Choose an element ej+1 of maximum weight
Xj+1 ← Xj ∪ {ej+1 }
j ←j+1
B ← Xj
Output B
We can characterize a matroid by using the greedy algorithm. Indeed, (I, E) is a
matroid if and only if the following conditions are verified
(I1) ∅ ∈ I
(I2) I ∈ I, I 0 ⊆ I ⇒ I 0 ∈ I
(G) For any function w : E → R, the greedy algorithm output a maximal set of I
of maximal weight.
3. Applications : combinatorial optimisation
3.1. Transversal matroid. Let S = {e1 , . . . , en } and let A = {A1 , . . . , Ak }, Ai ⊆ S,
n ≥ k. A transversal of A is a subset {ej1 , . . . , ejk } of S such that eji ∈ Ai (that is,
there exist a bijection between {ej1 , . . . , ejk } and {A1 , . . . , Ak }). A set X ⊆ S is said
to be a partial transversal of A if there exists {i1 , . . . , il } ⊆ {1, . . . , k} such that X is
a transversal of {Ai1 , . . . , Ail }.
Let G = (U, V ; E) be a bipartite graph formed from S = {s1 , . . . , sn } and A =
{A1 , . . . , Ak }, Ai ⊆ S where U = {u1 , . . . , un }, V = {v1 , . . . , vk } and two vertices
ui ∈ U and vj ∈ V are adjacents if and only if si ∈ Aj . We thus have that a set X is a
partial transversal of A if and only if there exists a matching of G = (U, V ; E) where
each edge of the matching has a vertex of U corresponding to one of the elements of
X, see Figure 3.
Example 3.1. Let E = {e1 , . . . , e6 } and A = {A1 , A2 , A3 , A4 } with A1 = {e1 , e2 , e6 },
A2 = {e3 , e4 , e5 , e6 }, A3 = {e2 , e3 } et A4 = {e2 , e4 , e6 }. Then, {e1 , e3 , e2 , e6 } is a
transversal of A and X = {e6 , e4 , e2 } is a partial transversal of A since X is a transversal
of {A1 , A2 , A3 }, see Figure 3.
It can be proved [9] that if E = {e1 , . . . , en } and if A = {A1 , . . . , Ak }, Ai ⊆ S then the
set of partial transversals of A is the set of independents of a matroid. Such a matroid
is called transversal.
6
JORGE RAMÍREZ ALFONSÍN
e1
e2
A1
e3
A2
e4
A3
e
A4
5
e6
Figure 3. Example of a transversal set
3.2. Assigment Problem. Let {Ti } be a set of works ordered by their importance
(priority) and let {Ei } a set of employees capable to do one or various of these works.
We suppose that the works will be done at the same time (and thus each employee can
do just one work each time). The problem is to assign the works to the employees in
a optimal way (maximizing the priorities).
The problem can be solved by applying the greedy algorithm to the transversal matroid
M where S = {Ti } and A = {A1 , . . . , Ak } with Ai the set of works for which employee
i is qualified. We notice that the maximal number of works that can be done at the
same time is equals to the biggest partial transversal of A with w : S → R the function
corresponding to the importance of the works.
Example 3.2. We have four works {t1 , t2 , t3 , t4 } to be done, each with a priority given
by the weights : w(t1 ) = 10, w(t2 ) = 3, w(t3 ) = 3 and w(t4 ) = 5. These works can be
done by three employees e1 , e2 and e3 : employee e1 is capable to do works t1 et t2 ,
employee e2 is capable to do works t2 et t3 , and employee e3 is capable to do work t4 .
Let M = (I, w) be the transversal matroid with I is the set of matchings of the bipartite
graph G = (U, V ; E) where U = {t1 , t2 , t3 , t4 }, V = {e1 , e2 , e3 } and ti ∈ U et ej ∈ V
are adjacent if and only if employee ej is capable to do work ti . By applying the greedy
algorithm to M it is obtained X0 = ∅, X1 = {t1 }, X2 = {t1 , t4 } and X3 = {t1 , t4 , t2 }.
4. Duality
Let M be a matroid on E and B the set of bases of M . Then,
B ∗ = {E\B | B ∈ B}
is the set of bases of a matroid on E.
The matroid on E having B ∗ as set of bases, denoted by M ∗ , is called dual of M . A
base in M ∗ is also called a cobase of M . It is clear that
r(M ∗ ) = |E| − rM and M ∗∗ = M .
Moreover the set I ∗ of independents of M ∗ is given by
I ∗ = {X | X ⊂ E such that there exist B ∈ B with X ∩ B = ∅},
and the rank function of M ∗ is given by
rM ∗ (X) = |X| + rM (E\X) − rM ,
for X ⊂ E.
The family C ∗ of circuits of M ∗ is the set of subsets D ⊂ E such that D ∩ B 6= ∅ for
any base B ∈ B and D is minimal by inclusion with this property.
THEORY OF MATROIDS AND APPLICATIONS
7
A circuit of M ∗ is also called cocircuit of M and a cocircuit of cardinality one is called
an isthme of M . It turns out that the cocircuits of a matroid M are the complements
of the hyperplans in the set of elements, i.e.
C ∗ = {E\H | H ∈ H}
where H is the set of hyperplanes of M .
Of course, the cocircuits of a matroid verify the same axioms as the circuits (see Section
2.2).
4.1. Graphic matroids. Let G = (V, E) be a graph. A cocycle, or cut, of G is a set
of edges joining the two parts of a partition of the set of vertices of G. A cocycle is
simple if and only if it is minimal by inclusion. Here, we only consider simple cocycles.
It is known that X is a cocycle of G = (E, V ) if and only if X a minimal subset of E
having a non empty intersection with each generating forest of G. Thus, if C(G)∗ is
the set of simple cocycles of a graph G then C(G)∗ is the set of circuits of a matroid of
E, called bond (or cocycle) matroid of G, denoted by B(G). We have
M ∗ (G) = B(G) and M (G) = B ∗ (G).
A natural question is the following one : is it true that if M is graphic then M ∗ is also
graphic ? For instance, if M = M (K4 ) where K4 is the complete graph on 4 vertices
then M ∗ is also graphic since M (K4 ) = M ∗ (K4 ). This property is not true in general,
for example, M ∗ (K5 ) is not graphic. However, this property is true if and only if G is
planar, that is when G admit a dual graph (see Figure 5 for an example). In this case,
we have
M ∗ (G) = M (G∗ ).
4.2. Minors. Let M be a matroid on E and A ⊂ E. Then, the set of independents of
a matroid on E\A is given by
{X ⊂ E\A | X est un independent of M }
This matroid is obtained from M by deleting the elements of A. Such a matroid is
denoted by M \A = M |E\A . In order to give the elements M \A, we might also use the
notation M (E\A). Moreover, the circuits of M \A are the circuits of M contained in
E\A and for X ⊂ E\A, we have rM \A (X) = rM (X).
Let M be a matroid on E, A ⊂ E and X ⊂ E\A. The following properties are
equivalents.
(i) There exist a base B of M |A such that X ∪ B is independent
(ii) For any base B of M |A the set X ∪ B is independent.
(iii) The set of independents of a matroid on E\A is
{X ⊂ E\A | there exists a base B of M |A such that X ∪ B is independent in M }
This matroid is obtained from M by contracting the elements of A. Such matroid is
denoted by M/A. Moreover, the circuits of M/A are the nonempty sets minimals by
inclusion of the form C\A for C circuit of M . For X ⊂ E\A, we have rM/A (X) =
rM (X ∪ A) − rM (A).
In the case when A consist of one element A = {e} for e ∈ E, we simplify notation by
M \e et M/e. The matroids M \e are M/e are sometimes called two principal minors
of M defined by e. Notice that M \e = M/e when e is either an isthme or a loop.
8
JORGE RAMÍREZ ALFONSÍN
1
4
1
4
3
2
3
5
2
(a)
1
4
2
3
(b)
(c)
Figure 4. (a) G; (b) G \ 5 and (c) G/5.
In a general way, a minor of a matroid M is any matroid obtained from M by a
sequence of deletions and contractions.
The operations deletion and contraction are associative and comutative, that is,
(M \A)\A0 = M \(A ∪ A0 ),
(M/A)/A0 = M/(A ∪ A0 )
and
(M \A)/A0 = (M/A0 )\A.
So, any minor of M is of the form
M \A/B = M/B\A,
for A, B ⊂ E disjoint.
The operations deletion and contraction are duals, that is,
(M \A)∗ = (M ∗ )/A and (M/A)∗ = (M ∗ )\A.
Example 4.1. Let r and n be integers 0 ≤ r ≤ n and E = {1, . . . , n}. Then, I = Er
the set of r-sets of E is the set of independents of a matroid, denoted by Un,r and called
uniform matroid. Let T ⊆ E with |T | = t. then,
Un−t,n−t if n ≥ t ≥ n − r,
Un,r \T =
Un−t,r if t < n − r.
We have
∗
\T
Un,r
= Un,n−r \T =
Un−t,n−t if n ≥ t ≥ r,
Un−r,n−t if t < r,
and thus
∗
Un,r /T = (Un,n−r \T ) =
(Un−t,n−t )∗ = U0,n−t if n ≥ t ≥ r,
(Un−r,n−t )∗ = Ur−t,n−t if t < r.
Example 4.2. In the graphic case, the notions of minors of the associated matroid
coincide with the usual notions on graphs. More precisely, let G = (V, E) be a graph
and let T ⊆ E. It can be verified that M (G) \ T = M (G \ T ) and M (G)/T = M (G/T ),
see Figure 4.
THEORY OF MATROIDS AND APPLICATIONS
6’
2’
3’
1’
2
2’
5’
3
4’
5
3’
1’
2
6
3
1
4
1
9
4’
5’
5
4
(b)
(a)
Figure 5. (a) G and its dual G∗ ; (b) G∗ \ 60 and (G∗ \ 6)∗ = G/6.
Therefore, we have that any minor of a graphic matroid is graphic. This property is not
true for other classes of matroids, for instance, it is not true for transversal matroids.
Figure 5 illustrate a planar graph, its dual and the deleting and contraction operation
of an edge.
5. Representable matroids
Let F be a field, d ≥ 1 an integer, E a finite set and V = (ve )e∈E a family of vectors of
Fd with index on E. We have already seen that
I = {X ⊂ E | the vectors ve , e ∈ X, are linearly independent on F}
is the set of independents of a matroid on E, called F-representables, or representable
over F.
Example 5.1. Matroid of Fano F7 . Figure 6. The matroid F7 is the finite projective
plan of order m = 2. By the classical formula [6] it has m2 + m + 1 = 7 elements. It can
be verified that F7 is the matroid of linear dependances on Z2 of 7 non zero vectors of
Z32 , that is the linear dependences on Z2 of the columns of matrix B. On the contrary,
this matroid is not representable over R.
 1
1

0
B=
0
2
0
1
0
3
0
0
1
4
0
1
1
5
1
0
1
6
1
1
0
7
1
1 
1
Remark 5.2. If M is defined by the vectors given by the columns (I | A) of size r × n,
where Ir is the identity r × r (their columns is an arbitray base on E) and A is the
matrix r × (n − r), then the dual matroid M ∗ is defined by the vector columns of the
matrix (−t A | In−r ) where In−r is the identity (n−r)×(n−r) and t A the transpose of A.
Some authors call M ∗ the orthogonal matroid of M since the duality for representable
matroids is a generalization of the notion of orthogonality in vector spaces. Indeed, let
V be a subspace of FE . We recall P
that the orthogonal espace V ⊥ is defined from the
canonical interior product hu, vi = e∈E u(e)v(e) by
V ⊥ = {v ∈ FE | hu, vi = 0 for any u ∈ V }.
It is known [9, Proposition 2.2.23] that the orthogonal espace generated by the column
vectors of (I | A) is given by the espace generated by the column vectors of (−t A | In−r ).
10
JORGE RAMÍREZ ALFONSÍN
3
5
1
7
6
4
2
Figure 6. Affine representation of Fano matroid F7 in the plane. F7
is a rank 3 matroid with 7 elements where three form a base if and
only if they are not on the same line. The line 456 is represented by a
circle. It is representable over Z2 (represented by matrix B), but it is
not representable over R.
.
By Remark 5.2 we have that the dual of a representable matroid M in a field F is
also representable in F. We also have that any minor of a representable over a field
F is also representable over F. Indeed, the deletion of an element is just the deletion
of the corresponding column in the matrix, and thus still representable in F. For
the contraction, we can simply apply deletion and contraction, that is, contracting an
element comes down to take the dual matroid (that is representable in F) delete an
element (obtaining a representable matroid in F) and then take the its dual (that is
again representable in F).
The property to be representable in a field F is thus kept by taking minors. For any
field F there exist a list of excluded minors, that is, not representable matroids in F
but where any proper minor is representable over F. The determination of a list of
excluded minors for F constitute a characterization of representable matroids over F
: a matroid is representable over F if and only if none of its minors is on the list of
excluded minors over F.
• For F = R, the list of exclded minors is infinite [9], and it looks difficult to
determine it.
• For F = GF (2) = Z2 = Z/2Z : the list is reduced to one mtaroid U2,4 [13, 14].
• For F = GF (3) = Z3 = Z/3Z : the list has 4 matroids F7 F7∗ U2,5 U3,5 [7].
• For F = GF (4) : the list has 8 matroids described explicitely in [5].
A big open question is the follwoing one : ls the list of excluded minors finite for a
given finite field ?
We finally note that there exist nonrepresentable matroids (that is, over any field), a
classical example is the rank 3 matroid obtained from the configuration of Non-Pappus,
see Figure 7.
THEORY OF MATROIDS AND APPLICATIONS
11
Figure 7. Non-Pappus. The matroid is defined on the 9 represented
points. The points over the same line form a circuits. In the vectorial
case, the three points in the middle are necessarily aligned (Pappus Theorem). Here (Non-Pappus) we decide that this three points form a base
instead of a circuit. We still obtain a matroid but not representable over
any field (see Section 5).
6. Regular matroids
A matroid representable over any field F is called regular. For example, graphic matroids are regular (see Remark 2.6). In general, regular matroids are equivalent to
totally unimodular matrices1.
Since any minor of a regular matroid is regular (by the results given in Section 5), we
can characterize regular matroids by a list of excluded minors. By a theorem due to
Tutte [13, 14] this list is finite having 3 matroids : U2,4 , F7 et F7∗ .
7. Union of matroids
Let M1 , M2 , . . . , Mk k ≥ 2 be matroids on E, and let Ii be the set of independents of
Mi for i = 1, 2, . . . , k. Let us set
I = {X1 ∪ X2 ∪ · · · ∪ Xk | Xi ∈ Ii for i = 1, 2, . . . , k}
Nash-Williams [8] proved that I is the set of independents of a matroid on E, called
union of matroids (Mi )i=1,2,...,k and denoted M1 ∨ M2 ∨ · · · ∨ Mk .
The rank function of M1 ∨ M2 ∨ · · · ∨ Mk for any A ⊂ E is given by
!
i=k
X
rM1 ∨M2 ∨···∨Mk (A) = min
ri (X) + |A\X| .
X⊂A
i=1
The assumption that the matroids Mi are on the same set is by commodity. Without
lost of generality, we can always come back to this case. Indeed, if the matroids Mi are
k
S
fi on
on different sets Ei , we set E =
Ei . For i = 1, . . . , k, we define the matroid M
i=1
fi (Ei ) = Mi and r f (e) = 0 for e ∈ E\Ei , that is, we extend Mi
E by the conditions M
Mi
fi .
to E by adding loops. Clearly, the set I do not change while passing from Mi to M
One of a number of consequences of the union of matroids is the well-kwon intersection
Theorem due to Edmonds [4] stating that for any integer k, there exist X ∈ I1 ∩ I2
such that |X| ≥ k if and only if r1 (A) + r2 (E\A) ≥ k for any A ⊂ E.
1A
modular matrix is a square matrix with integers coefficients having determinant equals to −1
or 1. A totally modular matrix is a matrix with coefficients 0,1,-1 where the determinat of all square
submatrices are equal to 0,1 or -1.
12
JORGE RAMÍREZ ALFONSÍN
One of the motivations of this result yielded from the fact that the corresponding result
in terms of graphs has had proved some years earlier.
It is natural to consider the problem on the generalization of the intersection theorem
of 2 matroids to the intersection of k ≥ 2 matroids. Unfortunately, such a result is not
known and looks unlikely the existence of such solution. One reason is given by the
theory of complexity of algorithms. A decision problem is said to be in the class N P
(‘deterministic polynomial’) if its solution can be verified in polynomial time, that is,
by an algorithm that uses a polynomial number of steps. A typical exemple of a N P
problem is satisfiability.
The intersection problem of 3 matroids is N P -complete. On the contrary, there exist
a polynomial time algorithm for the intersection of matroids is polynomial (assuming
the existence of an oracle for independency).
8. Tutte Polynomial
The Tutte polynôme de Tutte of a matroid M is the generating function defined as
t(M ; x, y) =
X
(x − 1)r(E)−r(X) (y − 1)|X|−r(X) .
X⊆E
This polynomial was introduced by Tutte [15] for graphs and then generilized to matroids by Crapo [3]. Its many rich properties, including several equivalent definitions,
have led to an abundant literature in a continual development [2]. This is a good example of remarkable object in the theory of matroids. After presenting a few fundamental
general properties we will give a more specifique applications.
Example 8.1. Recall that U3,2 is the uniform rank 2 matroid on 3 elements. Then,
t(U3,2 ; x, y) = x2 + x + y.
8.1. General properties. Recall that a loop of a matroid M is a circuit of cardinality
one and that an isthme of M is an element which is in all bases of M . The Tutte
polynomial can be expressed recursively as

 t(M \ e; x, y) + t(M/e; x, y) if e is neither a loop nor a isthme,
xt(M \ e; x, y)
if e is an isthme,
t(M ; x, y) =
 yt(M/e; x, y)
if e is a loop.
We state some basic and numerative properties of Tutte polynomial.
(i) t(M ∗ ; x, y) = t(M ; y, x).
(ii) Let M1 and M2 two matroids on the sets E1 et E2 respectively with E1 ∩ E2 = ∅.
Then, t(M1 ⊕ M2 ; x, y) = t(M1 ; x, y) · t(M2 ; x, y).
(iii) t(M ; 2, 2) counts the number of subsets of E.
(iv) t(M ; 1, 1) counts the number of bases of M .
(v) t(M ; 2, 1) counts the number of independents of M .
(vi) t(M ; 1, 2) counts the number of generating sets of M .
The Tutte polynomial appears in many counting problems in graph theory, matroids
and even in mechanical statistics.
THEORY OF MATROIDS AND APPLICATIONS
13
8.2. Chromatic Polynomial. Let G = (V, E) be a graph and let λ be a positive
integer. A λ-coloring of G is a function φ : V −→ {1, . . . , λ}. The coloring is said to be
good if for any edge {u, v} ∈ E(G), φ(u) 6= φ(v). Let χ(G, λ) be the number of good
λ-colorings of G.
The following result can be proved by using the inclusion-exclusion principle: if G =
(V, E) is a graph and λ is a positive integer, then,
X
χ(G, λ) =
(−1)|X| λω(G[X]) ,
X⊆E
where ω(G[X]) denote the number of connected components of the subgraphe induced
by X.
The chromatic polynomial was introduced by Birkhoff [1] as a tool to attack the four
color problem. Indeed, if for a planar graph G we have χ(G, 4) > 0 then G admit a good
4-coloring. The chromatic polynomial is essentially an evaluation of the Tutte polynomial of M (G). Indeed, if G = (V, E) is a graph with ω(G) connected components,
then
χ(G, λ) = λω(G) (−1)|V (G)|−ω(G) t(M (G); 1 − λ, 0).
8.3. Chromatic polynomial with defect. Let G be a graph, we define the chromatic
polynomial with defect B(G, λ, s) as
X
B(G, λ, s) =
bi (G, λ)si
i
, where bi (G, λ) denote the number λ-colorings of G with exactly i edges having the
their extremes with the same color. We have
s+λ−1
ω(G)
r(G)
B(G, λ, s) = λ
(s − 1) t M (G);
,s .
s−1
8.4. Flow Polynomial. Let G be an oriented graph (each edge has a positive end and
+
−
a negative end). Let S ⊂ V (G) and let ωG
(S) (resp. ωG
(S)) the set of edges of G
with positive ends (resp. negative ends) in S and negative ends (resp. positive ends) in
V \ S. Let H be an abelian group (with additive notation). A H-flow in G is a function
φ : E(G) −→ H such that the sum of weights of the edges getting into v is equal to
the sum of the weights of the edges getting out of v for all v ∈ V (G). A H-flow of G
is called nonwhere zero if φ(e) 6= 0 for all e ∈ E(G).
Let fλ (G) be the number of nonwhere zero H-flow of G. fλ (G) is called flow polynomial
of G. If G is connected and if H a finite abelian group of order λ, then
fλ (G) = (−1)|E(G)|−r(E) t(M (G); 0, 1 − λ).
The notion of nonwhere zero k-flow can be seen as the dual of good colorings. Indeed,
if G is a connected planar graph, then
χ(G, λ) = λFλ (G∗ ).
8.5. Ehrhart polynomial. The theory of Ehrhart was interested in counting the
number of integer points lying inside a polytope. We say that a polytope is integral
if all their vertices have integer coordinates. Given an integer polytope P , Ehrhart
studied the function iP that counts the number of integer points inside the P dilated
by a factor t, that is
14
JORGE RAMÍREZ ALFONSÍN
iP : N −→ N∗
t 7→ |tP ∩ Zd |
Ehrhart proved that the function iP is a polynomial on t of degree d,
iP (t) = cd td + cd−1 td−1 + · · · + c1 t + c0 .
iP (t) is called Ehrhart polynomial. The coefficients ci gives information about the
polytope P . Par example, cd is equals to V ol(P ) (the volume of P ), cd−1 is equals to
V ol(∂(P )/2) where ∂(P ) is the surface of P , and c0 = 1 is Euler’s characteristic of P .
All other coefficients remains a mystery.
Recall that the Minkowski sum of two sets A and B of Rd is
A + B = {a + b|a ∈ A, b ∈ B}.
Let V = {v1 , . . . , vk } be a finite set of elements of Rd . A zonotope generated by V ,
denoted by Z(A), is the polytope formed by the Minkowski sum of line segments
Z(A) = {α1 + · · · + αk |αi ∈ [−vi , vi ]}.
Let M be a regular matroid and let A one of its representation by a totally unimodular
matrix. Then, the Ehrhart polynomial associeted to zonotope Z(A) is given [12] by
1
r(M )
iZ(A) (q) = q
t M; 1 + , 1 .
q
8.6. Acyclic and totally cyclic orientations. Let G = (V, E) be a connected graph.
An orientation of G is an orientation of the edges of G. We say that an orientation is
acyclic if the oriented graph has not an oriented cycle (i.e., a cycle where the oriented
edges are all clockwise or anticlockwies). A classical result is due to Stanley [11] states
that the number of acyclic orientations a(G) of a graph G = (V, E) is
a(G) = (−1)|V (G)| χ(G, −1) = t(M (G); 2, 0).
An orientation of a graph G is called totally cyclic if every directed edge lies in at
least one directed cycle. One can show that this is equivalent to the condition that
the orientation on each connected component of G is strongly connected : for every
x, y ∈ V (G) in the same connected component of G there exist directed paths both
from x to y and from y to x. It turns out that the number of totally cyclic orientations
a∗ (G) of a connected graph G is :
a∗ (G) = a∗ (M (G)) = a(M (G)∗ ) = t(M (G); 0, 2).
References
[1] G.D. Birkhoff, A determinant formula for the number of ways of coloring a map, Annals of
Math. 14, 42-46 (1912-13).
[2] T. Brylawski, J. Oxley, The Tutte polynomial and its applications, Matroid Applications (ed.
N. White), Cambridge University Press, 123-225 (1992).
[3] H.H. Crapo, The Tutte polynomial, Aequationes Mathematicae 3, 211-229 (1969).
[4] J. Edmonds, Lehman’s switching game and a theorem of Tutte and Nash-Williams, Res. Nat.
Bur. Standards Sect. B, 69 73-77 (1965).
[5] J.F. Geeelen, A.M.H. Gerards, A. Kapoor, The excluded minors for GF(4)-representable matroids, J. Comb. Th. Ser. B 79, 247-299 (2000).
[6] D.R. Hughes, F.Piper, Projective Planes, Springer-Verlag, New York, (1973).
THEORY OF MATROIDS AND APPLICATIONS
15
[7] J. Kahn, P. Seymour, On forbidden minors for GF(3), Proc. Amer. Math. Soc., 102, 437440(1988).
[8] C. St. A. Nash-Williams, An application of matroids to graph theory, Theory of graphs (Int.
Sympos. Rome), Dunod Paris, 363-365 (1966).
[9] J. Oxley, Matroid Theory, Oxford University Press, (1992).
[10] P. Seymour, Decomposition of regular matroid, J. Comb. Th. Ser. B 28, 305-359 (1980).
[11] R.P. Stanley, Acyclic orientations of graphs, Disc. Math. 5, 171-178 (1973).
[12] R.P. Stanley, A zonotope associated with graphical degree sequences, DIMACS Series in Discrete
Mathematics and Theoretical Computer Science 4, 555-570 (1991).
[13] W.T. Tutte, A homotopy theorem for matroids I, Trans. Amer. Math. Soc. 88, 144-160 (1958).
[14] W.T. Tutte, A homotopy theorem for matroids II, Trans. Amer. Math. Soc. 88, 161-174 (1958).
[15] W.T. Tutte, A contribution to the theory of chromatic polynomials, Can. J. Math. 6, 80-91
(1954).
[16] H. Whitney, On the abstract properties of linear dependence, Amer. J. Math. 57, 509-533
(1935).
I3M, Université Montpellier 2, Place Eugène Bataillon, 34095 Montpellier, France
E-mail address: [email protected]